feat: Update hashbrown to instantiate less llvm IR

[Marwes]

Includes rust-lang/hashbrown#204 and rust-lang/hashbrown#205 (not yet merged) which both serve to reduce the amount of IR generated for hashmaps.

Inspired by the llvm-lines data gathered in #76680 (cc @Julian-Wollersberger)

[rust-highfive]

r? @Mark-Simulacrum

(rust_highfive has picked a reviewer for you, use r? to override)

[Marwes]

Prompted by rust-lang/hashbrown#205 (comment), this could use a perf run.

[jonas-schievink]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit c2c7c50ccb799451b06fb30a210bea2eb16ab059 with merge 4a35079c2dbc0f04dbb93febe7bf6f980ffb6a59...

[Marwes]

Had to update the the git dependency to point to the branch as it should, and fix the branch as hashbrown's ci didn't test the raw feature which rustc uses.

[lcnr]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit 534f867a487cbdad97a1dd97cf7552b96e87016a with merge e035000c1a2a0b031b1862e7b9b1d8f59efd7e7a...

[bors]

☀️ Try build successful - checks-actions, checks-azure
Build commit: e035000c1a2a0b031b1862e7b9b1d8f59efd7e7a (e035000c1a2a0b031b1862e7b9b1d8f59efd7e7a)

[rust-timer]

Queued e035000c1a2a0b031b1862e7b9b1d8f59efd7e7a with parent 62bfcfd, future comparison URL.

[Julian-Wollersberger]

Nice work! Here are some measurements of this PR.
(I recently wrote an entry in rustc-dev-guide how to do these measurements with a rustc debug build, to safe time.)

Sum of hashbrown lines before: 1 615 640
Sum of hashbrown lines after: 1 077 708
That's -537 932 lines! (-33%)

The total difference is 589 850 lines (-2.9%).

Here I filtered for hashbrown:

# Before:
  Lines            Copies         Function name
  -----            ------         -------------
  20188433 (100%)  596578 (100%)  (TOTAL)
    188221 (0.9%)     781 (0.1%)  hashbrown::raw::RawTable<T>::rehash_in_place
    139799 (0.7%)     781 (0.1%)  hashbrown::raw::RawTable<T>::resize
     89815 (0.4%)    1562 (0.3%)  hashbrown::raw::RawTable<T>::rehash_in_place::{{closure}}
     85074 (0.4%)    1289 (0.2%)  hashbrown::raw::RawTable<T>::find
     73140 (0.4%)     636 (0.1%)  hashbrown::raw::RawTable<T>::fallible_with_capacity

# After:
  19598583 (100%)  583746 (100%)  (TOTAL)
     89676 (0.5%)     636 (0.1%)  hashbrown::raw::RawTable<T>::rehash_in_place
     85860 (0.4%)     636 (0.1%)  hashbrown::raw::RawTable<T>::resize
     66576 (0.3%)     552 (0.1%)  hashbrown::raw::RawTable<T>::insert
     63210 (0.3%)     957 (0.2%)  hashbrown::raw::RawTable<T>::find
     58317 (0.3%)     536 (0.1%)  hashbrown::map::HashMap<K,V,S>::insert
     53424 (0.3%)     636 (0.1%)  hashbrown::raw::RawTableInner::fallible_with_capacity

[rust-timer]

Finished benchmarking try commit (e035000c1a2a0b031b1862e7b9b1d8f59efd7e7a): comparison url.

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. Please note that if the perf results are neutral, you should likely undo the rollup=never given below by specifying rollup- to bors.

Importantly, though, if the results of this run are non-neutral do not roll this PR up -- it will mask other regressions or improvements in the roll up.

@bors rollup=never

[Marwes]

Improvements in compiling rustc itself but there does seem to be a slight loss in performance when using the hashmap itself (I assume). I suspect it is the iteration code that got the slowdown so I will see if I can't tweak it a bit.

[Marwes]

Perf should be more stable now, another perf would be nice!

Not sure what is going on with the debug info providers yet, the changes in hashbrown forces an extra .table indirection but I don't seem to have added that correctly.

[lcnr]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit bf69d26632b6015f2b390f69cb2fc018341c571e with merge ce08d801f6fb1e045de36dbd8733c13eda8c0e37...

[bors]

☀️ Try build successful - checks-actions, checks-azure
Build commit: ce08d801f6fb1e045de36dbd8733c13eda8c0e37 (ce08d801f6fb1e045de36dbd8733c13eda8c0e37)

[rust-timer]

Queued ce08d801f6fb1e045de36dbd8733c13eda8c0e37 with parent 08e2d46, future comparison URL.

[bors]

⌛ Testing commit 81c9a02 with merge 547d71f63137d7c0fc97a5b627596cbfec5c3720...

[rust-log-analyzer]

A job failed! Check out the build log: (web) (plain)

Click to see the possible cause of the failure (guessed by this bot)

 19  516M   19  102M    0     0  24.4M      0  0:00:21  0:00:04  0:00:17 24.4M
 20  516M   20  105M    0     0  24.4M      0  0:00:21  0:00:04  0:00:17 24.7M
curl: (56) OpenSSL SSL_read: Connection was reset, errno 10054

gzip: stdin: unexpected end of file
tar: Unexpected EOF in archive
tar: Unexpected EOF in archive
tar: Error is not recoverable: exiting now
##[error]Process completed with exit code 2.
[command]"C:\Program Files\Git\bin\git.exe" version
git version 2.30.1.windows.1
[command]"C:\Program Files\Git\bin\git.exe" config --local --name-only --get-regexp core\.sshCommand
[command]"C:\Program Files\Git\bin\git.exe" submodule foreach --recursive "git config --local --name-only --get-regexp 'core\.sshCommand' && git config --local --unset-all 'core.sshCommand' || :"
[command]"C:\Program Files\Git\bin\git.exe" submodule foreach --recursive "git config --local --name-only --get-regexp 'core\.sshCommand' && git config --local --unset-all 'core.sshCommand' || :"
[command]"C:\Program Files\Git\bin\git.exe" config --local --name-only --get-regexp http\.https\:\/\/github\.com\/\.extraheader
http.https://github.com/.extraheader
[command]"C:\Program Files\Git\bin\git.exe" config --local --unset-all http.https://github.com/.extraheader
[command]"C:\Program Files\Git\bin\git.exe" submodule foreach --recursive "git config --local --name-only --get-regexp 'http\.https\:\/\/github\.com\/\.extraheader' && git config --local --unset-all 'http.https://github.com/.extraheader' || :"
Cleaning up orphan processes
Terminate orphan process: pid (6704) (node)
Terminate orphan process: pid (5984) (python)
Terminate orphan process: pid (1236) (vctip)

[bors]

💔 Test failed - checks-actions

[Amanieu]

@bors retry

[bors]

⌛ Testing commit 81c9a02 with merge 0464f63...

[bors]

☀️ Test successful - checks-actions
Approved by: Amanieu
Pushing 0464f63 to master...

[rylev]

@Marwes @Amanieu As you saw in your perf runs, this regresses performance slightly in the eval_to_const_value_raw query in the ctfe-stress-4-debug benchmark. We might want to look into that and see if we can figure out why that might be. I'm unfamiliar with that query though. It was mentioned that iteration performance might be impacted negatively by this upgrade. Any thoughts on this?

[ghost]

The iteration thing was a red herring, so I can't quite say why it would regress in that particular case.